Hydropower station auxiliary equipment operation state analysis and early warning method

Abstract

The application discloses a hydropower station auxiliary equipment operation state analysis and early warning method, relates to the technical field of fault early warning, and early warning steps are as follows: S1, based on sensor collection equipment operation parameters, preprocessing to build a data set, real-time transmission to the background; S2, using a multidimensional analysis method to evaluate the equipment operation state, converting continuous sensor data into discrete symbol sequences, building a state transition probability matrix, combining with the physical constraints of the equipment, and identifying the abnormal mode of the equipment through the entropy value change of the symbol sequence; S3, based on the sound sensor, the sound wave signal during the equipment operation is collected, and a sound vibration feature matrix is constructed. In the application, the state evaluation adopts multidimensional analysis, the continuous data is discretized, the state transition matrix, the entropy value change and the physical constraints of the equipment are combined, the implicit abnormality is accurately identified, the missed judgment and misjudgment are reduced, the early warning, accurate diagnosis and efficient operation and maintenance of the equipment are realized as a whole, the failure rate is reduced, the service life of the equipment is prolonged, and the stable operation of the hydropower station is ensured.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of fault early warning, in particular to a hydropower station auxiliary equipment operation state analysis and early warning method. BACKGROUND

[0002] As the core infrastructure for clean energy production, the stable operation of a hydropower station is directly related to the power supply reliability and energy supply safety of a power system. Auxiliary equipment, as a key support system for the operation of a hydropower station unit, covers multiple subsystems such as a cooling system, a lubrication system, a speed regulation system and a sealing system, and undertakes the important function of ensuring the efficient and safe operation of the main unit. However, in the long-term high-load and complex working condition operation process, auxiliary equipment is prone to faults such as component wear, sealing failure, pipeline blockage and motor abnormality. If such faults cannot be identified and warned in a timely manner, the auxiliary equipment may be shut down, which may further cause the main unit to reduce load or be shut down in an emergency, resulting in huge economic losses and even safety accidents. The traditional method in the prior art usually judges whether the equipment is abnormal based on a fixed threshold, without considering the dynamic correlation of the equipment operation parameters and the physical constraint conditions. In addition, the existing state evaluation method directly uses continuous sensor data for judgment without effective discretization processing and pattern recognition, which is difficult to capture the subtle trends of parameter changes and is prone to false alarms or missed alarms. When the auxiliary equipment is abnormal, manual experience is relied on for fault diagnosis, and there is a lack of precise identification means based on data driving. To solve the above technical problems, the hydropower station auxiliary equipment operation state analysis and early warning method is provided. SUMMARY

[0003] To solve the above technical problems, the hydropower station auxiliary equipment operation state analysis and early warning method is provided.

[0004] To achieve the above purpose, the technical scheme adopted by the application is as follows: the hydropower station auxiliary equipment operation state analysis and early warning method, the early warning steps are as follows:

[0005] S1, based on the sensor collected equipment operation parameters, a data set is constructed after preprocessing and is transmitted to the background in real time;

[0006] S2, a multi-dimensional analysis method is used to evaluate the equipment operation state, continuous sensor data is converted into a discrete symbol sequence, a state transition probability matrix is constructed, and the equipment is identified for abnormal mode through the entropy value change of the symbol sequence in combination with the physical constraints of the equipment;

[0007] S3, based on the sound sensor collected sound signals during the operation of the equipment, a sound vibration feature matrix is constructed, micro-fault features are extracted based on a singular value decomposition joint algorithm, and the abnormal state equipment is identified for fault type and positioned;

[0008] S4, according to the fault diagnosis result, corresponding early warning information is sent according to the preset risk level, and the early warning information is displayed in real time based on the display screen, and human-computer interaction is carried out.

[0009] Preferably, the sensor in S1 includes a piezoelectric acceleration sensor, an infrared sensor and a diffused silicon sensor; after the sensor data is transmitted to the on-site DTU, extreme values are removed by the 3σ principle, and fluctuations are smoothed by a sliding window; finally, the data is standardized into a JSON format with metadata; the sampling frequency is 10-100 Hz; the standardization is processed by Z-score; the sliding window length is set to 100 sampling points, and the window moving step is 50 sampling points;

[0010] The preprocessed data is recorded and stored, and the transmission process is based on a 5G / 4G network.

[0011] The piezoelectric acceleration sensor collects the vibration acceleration of the equipment, with a range of 0-50g and a sampling rate of 10kHz; the vibration displacement range is 0-10mm; the piezoelectric acceleration sensor is deployed on the surface of the key moving parts of the equipment's bearings, motor shaft and gear box;

[0012] The infrared sensor is used to collect the temperature of the key components, with a range of -20℃-200℃; the collected temperature includes bearing seat temperature and oil pipe wall temperature; the infrared sensor is deployed in the heating component area of the motor winding, hydraulic system and electronic control module of the equipment;

[0013] The diffused silicon sensor is used to collect the oil pressure, with a range of 0-10MPa; the diffused silicon sensor is installed at the pressure liquid level monitoring position of the equipment's hydraulic pipeline, air pressure cavity and cooling system circuit;

[0014] The 3σ principle removes extreme values based on the natural law of normal distribution, and removes abnormal data deviating from the mean value by more than 3 times the standard deviation;

[0015] When the sliding window is smoothed, the window size is set to 100 sampling points, which is based on the matching relationship between the sensor sampling frequency and the equipment running period.

[0016] Preferably, in S2, the continuous sensor data is converted into a discrete symbol sequence by a symbol aggregation approximation method, specifically:

[0017] After the data obtained by the sensor is preprocessed, it is converted into continuous time series data, the total length of the data is determined as N, the time series data with a length of N is divided into L non-overlapping windows at a fixed interval, where L

[0018] wherein the preset number of symbols is set to 5, and is derived based on the Gaussian distribution characteristics of the equipment operation parameters;

[0019] The window division is based on: the value of L is 1 / 10 of N, and N is the total length of the time series data.

[0020] The interval division standard is according to the Gaussian distribution probability density distribution, symbol 1 corresponds to (-∞, -1.28σ), symbol 2 corresponds to (-1.28σ, -0.52σ), symbol 3 corresponds to (-0.52σ, 0.52σ), symbol 4 corresponds to (0.52σ, 1.28σ), and symbol 5 corresponds to (1.28σ, +∞).

[0021] Preferably, the state transition probability matrix construction in the S2 step constructs an m x m state transition probability matrix with a preset number of symbols, where m = 5, and the symbols are used as row and column indexes. The transition probability is calculated by sliding window statistics symbol transition frequency, P(i→j) = symbol i to j transition times / symbol i total appearance times, and the matrix initial value is 0, which is updated synchronously every 10 sampling periods.

[0022] Preferably, the step of identifying the equipment abnormal mode in the S2 step is:

[0023] In combination with the physical constraints of the equipment, the upper limit of the speed is ≤110% of the rated speed, the critical value of the bearing temperature is ≤85℃, and the physical constraints of the vibration frequency range are 20Hz-1000Hz, invalid symbol combinations are excluded;

[0024] Based on the normal operation data covering more than 3 typical working conditions, each working condition ≥5000 symbol samples, a weight factor is introduced to calculate the information entropy, a time delay analysis is added to calculate the transition entropy, and an entropy value baseline and a three-level fluctuation threshold are established with a 95% confidence interval;

[0025] In real-time monitoring, the sliding window calculates the symbol sequence entropy value and checks the physical constraints. If the entropy value exceeds the baseline threshold or violates the physical constraints, it is determined to be abnormal. Both conditions are met at the same time, and it is marked as a high confidence abnormality.

[0026] Preferably, when it is determined to be abnormal, the physical constraint quantitative standard is: the upper limit of the speed is ≤110% of the rated speed of the equipment, the critical value of the bearing temperature is ≤85℃, and the vibration frequency range is 20Hz-1000Hz. Symbol combinations exceeding this range are determined to be invalid. If either condition of the entropy value exceeding the baseline threshold or the sequence violating the physical constraints is met, it is determined to be abnormal. Both conditions are met at the same time, and it is marked as a high confidence abnormality.

[0027] The specific numerical values of the physical constraints are obtained based on the mechanical design specifications of the equipment, and are specifically:

[0028] Upper limit of rotating speed, the rotating speed of the water turbine auxiliary motor is less than or equal to 1500 r / min, based on the constraint of motor rated power and mechanical strength;

[0029] Temperature threshold, the temperature of the bearing seat is less than or equal to 85 DEG C, based on the thermodynamic constraint of the high-temperature limit of lubricating oil;

[0030] Vibration frequency range, the normal operation vibration frequency is concentrated in 50-200 Hz, which is consistent with the inherent frequency rule of rotating machinery;

[0031] The increase of information entropy corresponds to the uniformization of symbol distribution, which reflects the increase of uncertainty of the equipment operation state.

[0032] Preferably, the fault type identification and positioning in the S3 step is based on the data collected by the acoustic sensor. After deploying the acoustic sensor on the equipment, the sampling parameters are set according to the working conditions, the acoustic signals are collected and preprocessed, the initial features are obtained through multi-domain feature extraction, the core features are screened based on the standard of Pearson correlation coefficient greater than or equal to 0.7 and variance contribution degree greater than or equal to 85%, and the acoustic vibration feature matrix is constructed with the sampling samples and the core features as the rows and columns.

[0033] The acoustic sensor is deployed at the motor, gear box, pump group, hydraulic system and fan.

[0034] Preferably, the step of identifying and positioning the fault type of the abnormal state equipment in the S3 step is as follows:

[0035] The matrix is subjected to mean centering and singular value decomposition SVD, the first k largest singular values with cumulative energy contribution degree greater than or equal to 90% are screened, and the principal component feature vector is formed in combination with the corresponding left singular vector; at the same time, the original acoustic wave signal is subjected to 3-5 layer wavelet packet decomposition by using db4 wavelet base function, the energy proportion of each frequency band is calculated after excluding the frequency band with signal-to-noise ratio less than 3, and the principal component feature vector is fused to form a fusion feature matrix.

[0036] Preferably, the abnormal feature vectors exceeding the threshold of 95% confidence interval are screened by Mahalanobis distance, the singular value contribution rate and high frequency band energy proportion quantitative indicators are extracted, the micro-fault feature set is obtained, the K-Nearest Neighbor algorithm with k value of 3-7 is used based on the labeled fault feature library containing typical faults such as bearing wear and impeller blockage, the Euclidean distance is matched between the to-be-diagnosed sample and the library features, the fault type is output when the matching degree is greater than or equal to 85%, and the fault type is marked as unknown fault when the threshold is lower; the feature vector difference of different measuring points is analyzed combined with the acoustic sensor deployment position and the equipment structure map, and the fault positioning is realized, and the fault feature library supports online updating and incremental training.

[0037] Preferably, the early warning information in the S4 step is displayed by early warning level layering, the first level is a mild prompt, the second level is a yellow pop-up window with a monitoring curve, and the third level is a red background plus a bee sound; the man-machine interaction design covers the query, control and feedback process, and different levels open corresponding permissions; when the first level early warning, the corresponding parameter deviates from the normal range by 10%-15%, without shutdown risk; when the second level early warning, the corresponding deviation range is 15%-25%, which may affect the service life of the equipment; and when the third level early warning, the corresponding deviation range is greater than or equal to 25%, which has a shutdown risk.

[0038] Compared with the prior art, the present application has the beneficial effects that:

[0039] In the present application, state evaluation adopts multi-dimensional analysis, discretizes continuous data, combines state transition matrix, entropy value change and equipment physical constraints, accurately identifies implicit abnormalities and reduces missed and misjudged cases; fault positioning relies on sound sensors to capture micro-fault sound and vibration signals, extracts core features by singular value decomposition, realizes fault type identification and component accurate positioning, solves the pain points of traditional monitoring in finding early problems, and realizes real-time visualization and man-machine interaction with the display screen, helping operation and maintenance personnel to make efficient decisions, and overall realizing early warning, accurate diagnosis and efficient operation of the equipment, reducing the failure rate, prolonging the service life of the equipment and ensuring the stable operation of the hydropower station. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 The early warning step flowchart of the present application. DETAILED DESCRIPTION

[0041] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only as examples, and other obvious variants can be thought of by those skilled in the art.

[0042] Referring to Figure 1 As shown in the figure, the water power station auxiliary equipment operation state analysis early warning method, the early warning step is:

[0043] S1, based on piezoelectric acceleration sensor, infrared sensor, diffused silicon sensor and sound sensor, respectively collecting device vibration acceleration, key component temperature, fluid pressure and sound wave signal, constructing data set after pretreatment, including collection time stamp, sensor number, subsystem identification metadata, transmitting to the background in real time through 5G / 4G network;

[0044] S2, adopting a multi-dimensional analysis method to evaluate the equipment operation state, converting continuous sensor data into discrete symbol sequence by symbolic aggregation approximation method, presetting the number of symbols to be 5, constructing a state transition probability matrix, combining with the physical constraints of the equipment, the upper limit of the rotating speed is less than or equal to 1500r / min, and the bearing seat temperature is less than or equal to 85℃, and identifying the abnormal mode of the equipment through the entropy value change of the symbol sequence;

[0045] S3, based on the sound sensor collected sound wave signal, the sampling rate is 44.1kHz, the sound vibration feature matrix is constructed, the micro-fault features are extracted through singular value decomposition combined with wavelet packet decomposition algorithm, and the fault type of abnormal state equipment is recognized and positioned;

[0046] S4, according to the fault diagnosis result, corresponding early warning information is sent according to the preset three-level risk grade, and the early warning information is displayed in real time based on the display screen, and human-computer interaction is carried out.

[0047] The sensors in S1 include piezoelectric acceleration sensors, infrared sensors and diffused silicon sensors; after the sensor data is transmitted to the field DTU, the extreme values are removed by the 3σ principle, the sliding window is smoothed, and finally the data is standardized into JSON format with metadata; the sampling frequency is 10Hz-100Hz; the standardization is processed by Z-score; the sliding window length is set to 100 sampling points, and the window moving step is 50 sampling points;

[0048] The preprocessed data is recorded and stored, and the transmission process is based on 5G / 4G network;

[0049] The piezoelectric acceleration sensor collects the vibration acceleration of the equipment, the range is 0-50g, the sampling rate is 10kHz, the range of vibration displacement is 0-10mm, and the piezoelectric acceleration sensor is deployed on the surface of the key moving parts of the bearing, motor shaft and gear box of the equipment;

[0050] The infrared sensor is used to collect the temperature of the key components, the range is-20℃~200℃, and the collected temperature includes bearing seat temperature and oil pipe wall temperature; the infrared sensor is deployed in the heating component area of the motor winding, hydraulic system and electronic control module of the equipment;

[0051] The diffused silicon sensor is used to collect the oil pressure, the range is 0-10MPa; the diffused silicon sensor is installed at the pressure liquid level monitoring position of the hydraulic pipeline, air pressure cavity and cooling system loop of the equipment;

[0052] The 3σ principle removes extreme values based on the natural law of normal distribution, and removes abnormal data deviating from the mean value by more than 3 times the standard deviation;

[0053] When the sliding window is smoothed, the window size is set to 100 sampling points, which is obtained based on the matching relationship between the sensor sampling frequency and the equipment running period.

[0054] In S2, the continuous sensor data is converted into a discrete symbol sequence by symbol aggregation approximation method, specifically:

[0055] The data obtained by the sensor is preprocessed and converted into continuous time series data, the total length of the data is determined as N, the time series data with a length of N is divided into L non-overlapping windows at a fixed interval, wherein L < N, the window size is N / L, and the non-integer is completed by interpolation, the mean value of the data in each window is calculated, and a sequence composed of L mean values is used to replace the original data, the dimension is reduced from N to L, and the mean value sequence after dimension reduction is standardized. Based on the preset number of symbols, the interval is divided according to the Gaussian distribution probability density;

[0056] The preset number of symbols is 5, which is obtained based on the Gaussian distribution characteristics of the equipment operation parameters;

[0057] The window division is based on: the value of L is 1 / 10 of N, and N is the total length of the time series data;

[0058] The interval division standard is distributed according to the Gaussian distribution probability density, symbol 1 corresponds to (-∞, -1.28σ), symbol 2 corresponds to (-1.28σ, -0.52σ), symbol 3 corresponds to (-0.52σ, 0.52σ), symbol 4 corresponds to (0.52σ, 1.28σ), and symbol 5 corresponds to (1.28σ, +∞).

[0059] In the S2 step, the state transition probability matrix is constructed with the preset number of symbols to construct an m x m state transition probability matrix, wherein m = 5, the symbol is used as the row and column index, the transition probability is calculated by sliding window statistics symbol transition frequency, P(i→j)= symbol i to j transition times / symbol i total appearance times, the matrix initial value is 0, and it is updated synchronously every 10 sampling periods.

[0060] Preferably, the step of identifying the device abnormal mode in the S2 step is:

[0061] In combination with the physical constraints of the device and the entropy value change, the upper limit of the speed is determined to be ≤110% of the rated speed, the critical value of the bearing temperature is ≤85℃, and the physical constraints of the vibration frequency range 20Hz-1000Hz are excluded from the invalid symbol combination;

[0062] Based on the normal operation data covering more than three typical working conditions, each working condition ≥5000 symbol samples, a weight factor is introduced to calculate the information entropy, a time delay analysis is added to calculate the transition entropy, and an entropy value baseline and a three-level fluctuation threshold are established with a 95% confidence interval;

[0063] In real-time monitoring, the sliding window calculates the symbol sequence entropy value and checks the physical constraints. If the entropy value exceeds the baseline threshold or violates the physical constraints, it is determined to be abnormal. Both satisfy the high confidence degree of abnormality.

[0064] Preferably, when the abnormality is determined, the physical constraint quantitative standard is: the upper limit of the rotating speed is less than or equal to 110% of the rated rotating speed of the device, the bearing temperature critical value is less than or equal to 85 DEG C, the vibration frequency range is 20-1000 Hz, and the symbol combination beyond the range is determined to be invalid; the entropy value exceeds the baseline threshold or the sequence violates the physical constraint, and any condition is met, which is determined to be abnormal; when both conditions are met, it is marked as a high confidence abnormality;

[0065] The specific values of the physical constraints are obtained based on the mechanical design specifications of the device, and are as follows:

[0066] The upper limit of the rotating speed is less than or equal to 1500 r / min, which is based on the rated power of the motor and the mechanical strength constraint.

[0067] The temperature critical value is less than or equal to 85 DEG C, which is based on the thermodynamic constraint of the high-temperature limit of lubricating oil.

[0068] The vibration frequency range is concentrated in 50-200 Hz, which is consistent with the inherent frequency rule of rotating machinery.

[0069] The increase of information entropy corresponds to the uniformization of symbol distribution, which reflects the increase of uncertainty of the device operating state.

[0070] The fault type identification and positioning in the S3 step is based on the data collected by the acoustic sensor. After deploying the acoustic sensor on the device, the sampling parameters are set according to the working conditions, the acoustic signals are collected and preprocessed, the initial features are obtained through multi-domain feature extraction, and after Z-score standardization, the core features are screened based on the standards of Pearson correlation coefficient greater than or equal to 0.7 and variance contribution degree greater than or equal to 85%, and the acoustic vibration feature matrix is constructed with the sampling samples and the core features as the rows and columns.

[0071] The acoustic sensor is deployed at the motor, gear box, pump group, hydraulic system and fan.

[0072] The fault type identification and positioning step for the abnormal state device in the S3 step is as follows:

[0073] The matrix is mean-centered and singular value decomposed SVD, the first k largest singular values with cumulative energy contribution degree greater than or equal to 90% are screened, and the principal component feature vector is constructed in combination with the corresponding left singular vector; at the same time, the db4 wavelet basis function is used for 3-5 layer wavelet packet decomposition on the original acoustic signal, the energy proportion of each frequency band is calculated after excluding the frequency band with signal-to-noise ratio less than 3, and the principal component feature vector is fused to form a fusion feature matrix.

[0074] The abnormal feature vector exceeding the 95% confidence interval threshold is screened by Mahalanobis distance, the singular value contribution rate and the high frequency band energy proportion are extracted as quantitative indicators, the micro-fault feature set is obtained, based on the labeled fault feature library containing typical faults such as bearing wear and impeller blockage, the K nearest neighbor algorithm with k value of 3-7 is adopted, the Euclidean distance is matched between the to-be-diagnosed sample and the library features, when the matching degree is greater than or equal to 85%, the fault type is output, and when the threshold is lower, it is marked as unknown fault; the difference of feature vectors of different measuring points is analyzed combined with the deployment position of the acoustic sensor and the equipment structure map, the fault positioning is realized, and the fault feature library supports online updating and incremental training.

[0075] The early warning information in the S4 step is displayed in a hierarchical manner according to the early warning level, the first level is a mild prompt, the second level is a yellow pop-up window with a monitoring curve, and the third level is a red background with a buzzing sound; the human-computer interaction design covers the query, control and feedback process, and different levels open corresponding permissions; when the first level warning occurs, the corresponding parameter deviates from the normal range by 10%-15%, and there is no shutdown risk; when the second level warning occurs, the corresponding deviation range is 15%-25%, which may affect the equipment life; when the third level warning occurs, the corresponding deviation range is ≥25%, which has a shutdown risk.

[0076] Data acquisition and preprocessing stage

[0077] In the data acquisition and preprocessing stage, a comprehensive monitoring network is constructed by using piezoelectric acceleration sensors, infrared sensors, diffused silicon sensors and acoustic sensors dedicated to fault identification, and the details are as follows:

[0078] The specific method of Z-score standardization is as follows: the original data collected by the sensor (including vibration acceleration, temperature, oil pressure) is subtracted from the mean value of the parameter under normal working conditions, and then divided by the standard deviation under normal working conditions, where the mean value and standard deviation under normal working conditions are obtained by statistical analysis of three typical working conditions, each condition not less than 5000 samples, such as the mean value of bearing temperature is usually 45℃, the standard deviation is 8℃, the mean value of vibration acceleration is 5g, and the standard deviation is 1.2g. After such processing, the data will conform to the standard normal distribution;

[0079] The sliding window smoothing calculation method is as follows: the moving average method is used for smoothing, taking 100 sampling points as a window, and taking the average value of all original data of the sensor in the window as the smoothed data point corresponding to the window. This method can effectively filter high-frequency noise, such as transient interference in hydraulic pipeline pressure fluctuation, while retaining the overall trend of parameter change;

[0080] The pre-processed sensor data (including vibration acceleration, temperature and oil pressure) in this stage is the only original data source for subsequent symbol aggregation approximation processing, state transition probability matrix construction and entropy calculation. The data after Z-score standardization processing eliminates the dimensional differences between different parameters, providing a unified benchmark for subsequent symbol interval division. The data analysis after sliding window smoothing reduces noise interference, ensuring that the converted symbol sequence can accurately reflect the device operating state changes. The data with metadata stored in JSON format can also provide a basis for subsequent fault location.

[0081] The weight factor value and basis, the weight factor is introduced in the information entropy calculation according to the importance of the parameter. The weight of the vibration parameter (acceleration, displacement) is set to 0.4, the weight of the temperature parameter is 0.3, and the weight of the oil pressure parameter is 0.3. The weight allocation is determined based on mechanical fault statistical data. Because the proportion of faults caused by vibration abnormalities in actual operation is more than 60%, and the proportion of faults caused by temperature and oil pressure abnormalities is about 20% respectively, such weight setting can make the entropy value more prominent the influence of key parameters.

[0082] The time delay parameter value and basis, the time delay of transition entropy calculation is set to 1 to 3 sampling periods. For example, when the sampling frequency is 100 Hz, the time delay corresponds to 10 milliseconds, 20 milliseconds or 30 milliseconds. This time delay range is determined based on the device running period characteristics to assist in capturing the lag characteristics of symbol transition in the early stage of failure when the device speed is 1500 revolutions per minute.

[0083] Entropy baseline and three-level fluctuation threshold, the information entropy baseline is obtained by statistical analysis of three typical working conditions, each working condition with not less than 5000 symbol samples. The value range is between 1.2 and 1.8. The corresponding three-level fluctuation thresholds are 1.8 to 2.2 for slight deviation, 2.2 to 2.6 for moderate deviation, and not less than 2.6 for severe deviation. The transition entropy baseline value range is 0.8 to 1.3, and the three-level fluctuation thresholds are 1.3 to 1.7, 1.7 to 2.1, and not less than 2.1. These thresholds are determined by 95% confidence interval, which can cover the entropy value fluctuation range of 95% normal running samples.

[0084] Symbol sequence probability calculation supplement, the appearance probability of a symbol is obtained by dividing the number of symbol appearances by the total number of symbols, where the total number of symbols is one-tenth of the total length of the original time series data, and the number of symbol appearances is based on the pre-processed sensor time series data.

[0085] Information entropy calculation, the calculation of information entropy needs to first count the probability of occurrence of each symbol, then multiply the probability of occurrence of each symbol by its logarithm, multiply the weight of the corresponding parameter, take the negative number, and finally sum all the results. The symbol occurrence probability is obtained by counting the symbol sequence converted from the data of the piezoelectric acceleration sensor (vibration), infrared sensor (temperature), and diffused silicon sensor (oil pressure). The weight factor also corresponds to the parameters collected by the above three types of sensors.

[0086] Transition entropy calculation, transition entropy calculation needs to count the joint transition probability of the current symbol and the symbol after the time delay, as well as the marginal probability of the symbol after the time delay. The logarithm of the ratio of the joint transition probability and the marginal probability is taken, and then summed to obtain the joint transition probability. The current symbol and the symbol after the time delay are both from the symbol sequence converted from the sensor data. The joint transition probability and the marginal probability are obtained based on the time series characteristics of the sensor data.

[0087] State transition probability calculation, the transition probability from one symbol to another symbol is obtained by dividing the number of transitions of the symbol combination by the total number of occurrences of the first symbol. The number of transitions of the symbol combination and the total number of occurrences of the first symbol are both dependent on the real-time collection data of the vibration, temperature, and oil pressure sensors converted into symbol sequences.

[0088] The specific values of the physical constraints are as follows: The upper limit of the speed of the water turbine auxiliary motor is set to 1500 revolutions per minute, because its rated speed is usually 1360 revolutions per minute, and 110% of the rated speed is about 1496 revolutions per minute, which is rounded to 1500 revolutions per minute. The upper limit of the bearing seat temperature is set to 85℃, based on the fact that the high temperature limit of commonly used 32# anti-wear hydraulic oil is 90℃, with a 5℃ safety margin. The vibration frequency range is set to 20 Hz to 1000 Hz, because the vibration frequency of the device during normal operation is concentrated between 50 Hz and 200 Hz, and exceeding 1000 Hz is mostly component impact failure, and below 20 Hz is mostly environmental interference signal.

[0089] Sound sensor sampling parameter quantization, the sampling frequency of low-speed equipment (speed not exceeding 1500 revolutions per minute) is set to 2 kHz, the sampling time is 10 seconds, and the preprocessing uses 50 Hz high-pass filtering plus trend removal processing. The sampling frequency of medium-high speed equipment (speed between 1500 revolutions per minute and 3000 revolutions per minute) is 5 to 10 kHz, the sampling time is 8 seconds, and the preprocessing uses 100 Hz high-pass filtering plus trend removal processing. The sampling frequency of high-speed equipment (speed not less than 3000 revolutions per minute) is 10 to 20 kHz, the sampling time is 5 seconds, and the preprocessing uses 200 Hz high-pass filtering plus trend removal processing. The purpose of high-pass filtering is to filter out noise such as water flow and environment concentrated below 50 Hz, and trend removal processing is used to eliminate the baseline drift caused by sensor installation deviation.

[0090] The characteristic quantification index of typical faults is based on statistical data of no less than 1000 samples for each fault. The cumulative contribution rate of the top 3 singular values of the bearing wear fault is not less than 60%, the high-frequency band energy ratio under a sampling frequency of 10 kHz is not less than 30%, and the Pearson correlation coefficient between the vibration acceleration and the sound signal is not less than 0.82. The cumulative contribution rate of the top 3 singular values of the impeller blockage fault is not less than 50%, the high-frequency band energy ratio under a sampling frequency of 5 kHz is not less than 25%, and the Pearson correlation coefficient between the oil pressure and the sound signal is not less than 0.78. The cumulative contribution rate of the top 3 singular values of the motor imbalance fault is not less than 55%, the high-frequency band energy ratio under a sampling frequency of 8 kHz is not less than 28%, and the Pearson correlation coefficient between the vibration frequency and the sound signal is not less than 0.80. The cumulative contribution rate of the top 3 singular values of the seal leakage fault is not less than 45%, the high-frequency band energy ratio under a sampling frequency of 2 kHz is not less than 22%, and the Pearson correlation coefficient between the air pressure and the sound signal is not less than 0.75.

[0091] Mahalanobis distance threshold quantification. The Mahalanobis distance threshold is calculated based on the covariance matrix of the normal sample acoustic vibration feature matrix, and then the core feature number is used to query the chi-square distribution table to determine the threshold corresponding to the 95% confidence interval, which is 6.31.

[0092] K-Nearest Neighbor algorithm related parameters. The k value of the K-Nearest Neighbor algorithm is determined by 5-fold cross-validation. When k is 5, the validation set accuracy is the highest, reaching 92.3%, so the default k value is 5. The Euclidean distance matching threshold is set to not less than 85%. The fault diagnosis accuracy corresponding to this threshold is not less than 90%. This result is verified based on more than 1000 historical fault cases.

[0093] Singular value decomposition processing. The processing object of singular value decomposition is the acoustic vibration feature matrix constructed from acoustic sensor data. The rows of the matrix correspond to sampling samples, and the columns correspond to core features. Through decomposition, left singular vectors, singular value matrices, and right singular vectors can be obtained. The core features come from the multi-domain extraction of acoustic sensor data, including time domain, frequency domain, time-frequency domain, etc.

[0094] Wavelet packet decomposition frequency band energy ratio calculation. First, use db4 wavelet basis function to decompose the signal collected by the acoustic sensor for 3 to 5 layers. After decomposition, remove the frequency bands with a signal-to-noise ratio less than 3. Then calculate the sum of the squares of each frequency band coefficient (i.e., the energy of the frequency band). Divide the energy of a single frequency band by the total energy of all frequency bands and multiply by 100% to get the energy ratio of the frequency band. The high, medium, and low frequency band energy ratios for different fault types differ significantly, and this index is an important basis for distinguishing fault types.

[0095] Mahalanobis distance calculation, Mahalanobis distance is used to screen abnormal feature vectors, and the calculation needs to obtain the fusion feature vector of the sample to be diagnosed, and the mean and covariance matrix of the normal sample fusion feature vector, and the difference between the sample vector to be diagnosed and the mean vector is calculated, combined with the inverse matrix of the covariance matrix, when the calculation result exceeds 6.31, it is determined that the feature vector is abnormal, wherein the fusion feature vector, mean and covariance matrix are calculated based on the data collected by the acoustic sensor.

[0096] Euclidean distance matching, Euclidean distance is used to match the historical samples in the fault feature library with the sample to be diagnosed, and the difference between each index (including singular value contribution rate, frequency band energy proportion) in the micro-fault feature set of the sample to be diagnosed and the corresponding index of the historical sample is squared and summed, and then the square root is taken to obtain the distance value. When the matching degree corresponding to the distance value is not less than 85%, the corresponding fault type is output. The micro-fault feature set of the sample to be diagnosed and the historical sample feature set are both from the feature extraction and quantization of acoustic sensor data.

[0097] Pearson correlation coefficient screening, Pearson correlation coefficient is used to measure the correlation between different initial features (such as time domain peak value, frequency domain gravity center) extracted from acoustic sensor data. The calculation is obtained by dividing the sum of the product of the difference between each feature sample data and its mean value by the product of the standard deviations of the two features. The features with an absolute value of the correlation coefficient not less than 0.7 are retained to avoid feature redundancy and ensure the independence of the core features.

[0098] Variance contribution degree screening, variance contribution degree is used to measure the explanatory ability of a single feature to the original acoustic signal information. The calculation is obtained by dividing the variance of a single feature by the sum of the variances of all features and multiplying by 100%. The features with a cumulative variance contribution degree not less than 85% are selected to ensure that the retained core features can explain more than 85% of the information changes of the original acoustic signal.

[0099] Quantitative benchmark for parameter deviation from normal range, the normal range of the parameter is set to the mean value of the sensor data under normal working conditions plus or minus one standard deviation, for example, the normal range of bearing temperature is 45℃ plus or minus 8℃, i.e. between 37℃ and 53℃. The proportion of the parameter deviating from the normal range is obtained by dividing the absolute value of the difference between the real-time data collected by the sensor and the mean value under normal working conditions by the mean value under normal working conditions and multiplying by 100%.

[0100] The correlation of the early warning threshold value and the entropy value and the fault feature, the first level early warning corresponds to the parameter deviating from the normal range by 10% to 15%, at this time the information entropy is between 1.8 and 2.2, and there is no clear fault feature (the matching degree is lower than 60%); the second level early warning corresponds to the parameter deviating from the normal range by 15% to 25%, the information entropy is between 2.2 and 2.6, and the fault feature matching degree is between 60% and 85%; the third level early warning corresponds to the parameter deviating from the normal range by not less than 25%, the information entropy is not less than 2.6, and the fault feature matching degree is not less than 85%.

[0101] The early warning level is essentially a direct mapping of the severity of the sensor data deviating from the normal range, the first level light prompt corresponds to small amplitude fluctuation of the sensor data, such as temperature between 53 DEG C and 59 DEG C, not reaching the physical constraint threshold, the entropy value slightly increases, and there is no clear fault feature; the second level yellow pop-up prompt corresponds to moderate deviation of the sensor data, such as temperature between 59 DEG C and 66 DEG C, which may approach the physical constraint threshold, the entropy value moderately increases, and a suspected fault feature appears; the third level red early warning corresponds to serious deviation of the sensor data, such as temperature not less than 66 DEG C, which has violated the physical constraint or exists high confidence anomaly, the entropy value significantly increases, and the fault feature is clearly matched.

[0102] The "monitoring curve" in human-computer interaction is the visual presentation of real-time sensor data, such as vibration acceleration time series curve, temperature change curve, etc.; the "fault prompt" is generated based on the fault type identification result derived from the sensor data. The disposal result fed back by the staff through the interactive interface can also feed back the online update of the fault feature library, such as supplementing the sensor data samples of unknown faults.

[0103] The core input of all subsequent analysis steps is the data collected by the sensor, and the preprocessed data is the basis for symbol sequence construction and feature matrix generation. Without sensor data, symbol conversion and feature extraction cannot be performed; the entropy value, transition probability, Markov distance and other quantitative indicators are all indirect mappings of sensor data, and the numerical change essentially reflects the running state of the sensor monitoring parameter; the final result of fault identification and early warning is the conclusion obtained by comparing the sensor data with the preset threshold value and historical samples, and the accuracy and integrity of the sensor data directly determine the reliability of fault diagnosis and early warning.

[0104] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above examples, and the above examples and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing and early warning of the running state of auxiliary equipment of a hydropower station, characterized in that, The early warning steps are: S1, based on the sensor acquisition device operating parameters, pre-processing to build a dataset, real-time transmission to the background; the sensor includes a piezoelectric acceleration sensor, an infrared sensor and a diffused silicon sensor, the piezoelectric acceleration sensor collects the vibration acceleration of the equipment, the infrared sensor is used to collect the temperature of the key components, and the diffused silicon sensor is used to collect the oil pressure; S2, a multi-dimensional analysis method is used to evaluate the equipment operating state, the continuous sensor data is converted into a discrete symbol sequence, a state transition probability matrix is constructed, combined with the physical constraints of the equipment, the abnormal mode of the equipment is identified through the entropy value change of the symbol sequence; The step of identifying the abnormal mode of the equipment in the S2 step is: Combined with the physical constraints and the entropy value change, the upper limit of the speed is determined to be less than or equal to 110% of the rated speed, the critical value of the bearing temperature is less than or equal to 85℃, and the physical constraints of the vibration frequency range 20Hz-1000Hz are excluded from the invalid symbol combination; Based on the normal operation data covering more than 3 typical working conditions, each working condition ≥5000 symbol samples, a weight factor is introduced to calculate the information entropy, a time delay analysis is added to calculate the transfer entropy, and an entropy value baseline and a three-level fluctuation threshold are established with a 95% confidence interval; In real-time monitoring, the symbol sequence entropy value is calculated by a sliding window and the physical constraints are checked, if the entropy value exceeds the baseline threshold or violates the physical constraints, it is determined to be abnormal, and both conditions are met at the same time, which is marked as a high confidence abnormality; S3, based on the sound sensor, the sound wave signal of the equipment during operation is collected, a sound vibration feature matrix is constructed, and a singular value decomposition joint algorithm is used to extract micro-fault features, and the fault type of the abnormal state equipment is identified and located; S4, according to the fault diagnosis result, the corresponding early warning information is sent according to the preset risk level, and the early warning information is displayed in real time based on the display screen for human-computer interaction.

2. The method for analyzing and early warning of the operation state of auxiliary equipment of a hydropower station according to claim 1, characterized in that: After the sensor data is transmitted to the field DTU, the extreme values are removed by the 3σ principle, the sliding window is smoothed, and finally it is standardized to JSON format with metadata; the sampling frequency is 10Hz-100Hz; the standardization is processed by Z-score; the sliding window length is set to 100 sampling points, and the window moving step is 50 sampling points; The pre-processed data is recorded and stored, and the transmission process is based on 5G or 4G network; The range of the piezoelectric acceleration sensor is 0-50g, the sampling rate is 10kHz, and the range of the vibration displacement is 0-10mm; the piezoelectric acceleration sensor is deployed on the surface of the key moving parts of the bearings, motor shafts and gearboxes of the equipment; The range of the infrared sensor is -20℃~200℃, and the collected temperature includes bearing seat temperature and oil pipe wall temperature; the infrared sensor is deployed in the heating component areas of the motor winding, hydraulic system and electronic control module of the equipment; The range of the diffused silicon sensor is 0-10MPa; the diffused silicon sensor is installed at the pressure liquid level monitoring positions of the hydraulic pipeline, air pressure cavity and cooling system loop of the equipment; The 3σ principle removes the extreme values based on the normal distribution law, and removes the abnormal data deviating from the mean value by more than 3 times the standard deviation; When the sliding window is smoothed, the window size is set to 100 sampling points, which is based on the matching relationship between the sensor sampling frequency and the equipment operating cycle.

3. The method for analyzing and early warning of the operation state of auxiliary equipment of a hydropower station according to claim 1, characterized in that: The continuous sensor data is converted into a discrete symbol sequence in the S2 step, which is processed by a symbol aggregation approximation method, specifically: The data obtained by the sensor is preprocessed and converted into continuous time series data. The total length of the data is determined as N. The time series data with a length of N is divided into L non-overlapping windows at a fixed interval, where L < N, the window size is N / L, and if it is not an integer, it is completed by interpolation. The mean value of the data in each window is calculated, and a sequence composed of L mean values is used to replace the original data, reducing the dimension from N to L. The mean value sequence after dimension reduction is standardized. Based on the preset number of symbols, the intervals are divided according to the Gaussian distribution probability density; The preset number of symbols is set to 5, which is derived based on the Gaussian distribution characteristics of the device operating parameters; The window division is based on: L value is 1 / 10 of N, N is the total length of the time series data; The interval division standard is according to the Gaussian distribution probability density distribution, symbol 1 corresponds to (-∞, -1.28σ), symbol 2 corresponds to (-1.28σ, -0.52σ), symbol 3 corresponds to (-0.52σ, 0.52σ), symbol 4 corresponds to (0.52σ, 1.28σ), and symbol 5 corresponds to (1.28σ, +∞).

4. The method for analyzing and early warning of the operation state of auxiliary equipment of a hydropower station according to claim 1, characterized in that: The state transition probability matrix in the S2 step is constructed with the preset number of symbols to construct an m x m state transition probability matrix, where m = 5. The symbols are used as row and column indices. The transition probability is calculated by sliding window statistics symbol transition frequency, P(i→j)=symbol i to j transition times / symbol i total appearance times. The matrix initial value is 0, and it is updated synchronously every 10 sampling periods.

5. The method for analyzing and early warning of the operation state of auxiliary equipment of a hydropower station according to claim 1, characterized in that: The specific numerical values of the physical constraints are obtained based on the mechanical design specifications of the device, specifically: Upper limit of rotational speed, water turbine auxiliary motor speed ≤ 1500 r / min, based on motor rated power and mechanical strength constraints; Temperature critical value, bearing seat temperature ≤ 85℃, based on the thermodynamic constraints of high temperature resistance limit of lubricating oil; Vibration frequency range, normal operation vibration frequency concentrated in 50-200Hz, consistent with the inherent frequency law of rotating machinery vibration; Information entropy increase corresponds to uniform distribution of symbols, reflecting the increase of uncertainty of device operating state.

6. The method for analyzing and early warning of the operation state of auxiliary equipment of a hydropower station according to claim 1, characterized in that: The fault type identification and positioning in the S3 step is based on the data collected by the acoustic sensor. After deploying the acoustic sensor on the device, set the sampling parameters according to the working conditions, collect the acoustic signals and preprocess them, obtain the initial features through multi-domain feature extraction, and select the core features based on the standard of Pearson correlation coefficient ≥ 0.7 and variance contribution degree ≥ 85%. The acoustic vibration feature matrix is constructed with the sampling samples and core features as rows and columns; The acoustic sensor is deployed at the motor, gear box, pump group, hydraulic system and fan.

7. The method for analyzing and early warning of the operation state of auxiliary equipment of a hydropower station according to claim 6, characterized in that, The steps of fault type identification and positioning for abnormal state devices in the S3 step are: The matrix is mean-centered and singular value decomposition (SVD) is performed, the first k largest singular values with cumulative energy contribution degree greater than or equal to 90% are screened, and the corresponding left singular vectors are combined to form principal component feature vectors; meanwhile, the original acoustic signal is subjected to 3-5 layer wavelet packet decomposition by using a db4 wavelet base function, the energy proportion of each frequency band is calculated after eliminating the frequency band with a signal-to-noise ratio less than 3, and the principal component feature vectors are fused to form a fusion feature matrix.

8. The method for analyzing and early warning of the operation state of auxiliary equipment of a hydropower station according to claim 1, characterized in that: The early warning information in the S4 step is displayed in a hierarchical manner according to early warning levels, the first level is a mild prompt, the second level is a yellow pop-up window with a monitoring curve, and the third level is a red background with a buzzing sound; the man-machine interaction design covers the query, control and feedback processes, and the corresponding permissions are opened for different levels; when the first level early warning occurs, the corresponding parameters deviate from the normal range by 10%-15%, and there is no shutdown risk; when the second level early warning occurs, the corresponding deviation range is 15%-25%, which may affect the service life of the equipment; and when the third level early warning occurs, the corresponding deviation range is greater than or equal to 25%, which exists a shutdown risk.

Images